This paper proposes a novel Orthogonal frequency-division multiplexing (OFDM) system based on polynomial cancellation coded OFDM (PCC-OFDM). This proposed system can reduce peak-to-average power ratio (PAPR) by our neural phase rotator and it does not need any side information to transmit phase rotation factors. Moreover, this system can compensate the common phase error (CPE) by a proposed technique which allows estimating frequency offset at receiver. From numerical experiments, it is shown that our system can reduce PAPR and ICI at the same time and improve BER performance effectively.

This paper proposes a method (Group-Linking Method) that has control over the complexity of the sequential function to construct Finite Memory Machines with minimal order--the machines have the largest number of states based on their memory taps. Finding a machine with maximum number of states is a nontrivial problem because the total number of machines with memory order k is (256)2k-2, a pretty large number. Based on the analysis of Group-Linking Method, it is shown that the amount of data necessary to reconstruct an FMM is the set of strings not longer than the depth of the machine plus one, which is significantly less than that required for traditional greedy-based machine learning algorithm. Group-Linking Method provides a useful systematic way of generating unified benchmarks to evaluate the capability of machine learning techniques. One example is to test the learning capability of recurrent neural networks. The problem of encoding finite state machines with recurrent neural networks has been extensively explored. However, the great representation power of those networks does not guarantee the solution in terms of learning exists. Previous learning benchmarks are shown to be not rich enough structurally in term of solutions in weight space. This set of benchmarks with great expressive power can serve as a convenient framework in which to study the learning and computation capabilities of various network models. A fundamental understanding of the capabilities of these networks will allow users to be able to select the most appropriate model for a given application.

In this paper, we present a new method to enhance classification performance of a multiple classifier system by combining a boosting technique called AdaBoost.M2 and Kernel Discriminant Analysis (KDA). To reduce the dependency between classifier outputs and to speed up the learning, each classifier is trained in a different feature space, which is obtained by applying KDA to a small set of hard-to-classify training samples. The training of the system is conducted based on AdaBoost.M2, and the classifiers are implemented by Radial Basis Function networks. To perform KDA at every boosting round in a realistic time scale, a new kernel selection method based on the class separability measure is proposed. Furthermore, a new criterion of the training convergence is also proposed to acquire good classification performance with fewer boosting rounds. To evaluate the proposed method, several experiments are carried out using standard evaluation datasets. The experimental results demonstrate that the proposed method can select an optimal kernel parameter more efficiently than the conventional cross-validation method, and that the training of boosting classifiers is terminated with a fairly small number of rounds to attain good classification accuracy. For multi-class classification problems, the proposed method outperforms both Boosting Linear Discriminant Analysis (BLDA) and Radial-Basis Function Network (RBFN) with regard to the classification accuracy. On the other hand, the performance evaluation for 2-class problems shows that the advantage of the proposed BKDA against BLDA and RBFN depends on the datasets.

Skin tone detection in conditions where illuminate intensity and/or chromaticity can vary often comes with high computational time or low accuracy. Here a technique is presented integrating chromaticity and intensity normalization combined with a neural skin tone classification network to achieve robust classification faster than other approaches.

This paper presents a novel feature extraction algorithm based on particle swarms for processing hyperspectral imagery data. Particle swarm optimization, originally developed for global optimization over continuous spaces, is extended to deal with the problem of feature extraction. A formulation utilizing two swarms of particles was developed to optimize simultaneously a desired performance criterion and the number of selected features. Candidate feature sets were evaluated on a regression problem. Artificial neural networks were trained to construct linear and nonlinear models of chemical concentration of glucose in soybean crops. Experimental results utilizing real-world hyperspectral datasets demonstrate the viability of the method. The particle swarms-based approach presented superior performance in comparison with conventional feature extraction methods, on both linear and nonlinear models.

Over the years, many improvements and refinements to the backpropagation learning algorithm have been reported. In this paper, a new adaptive penalty-based learning extension for the backpropagation learning algorithm and its variants is proposed. The new method initially puts pressure on artificial neural networks in order to get all outputs for all training patterns into the correct half of the output range, instead of mainly focusing on minimizing the difference between the target and actual output values. The upper bound of the penalty values is also controlled. The technique is easy to implement and computationally inexpensive. In this study, the new approach is applied to the backpropagation learning algorithm as well as the RPROP learning algorithm. The superiority of the new proposed method is demonstrated though many simulations. By applying the extension, the percentage of successful runs can be greatly increased and the average number of epochs to convergence can be well reduced on various problem instances. The behavior of the penalty values during training is also analyzed and their active role within the learning process is confirmed.

We discuss a new design methodology of polynomial neural networks (PNN) in the framework of genetic algorithm (GA). The PNN is based on the ideas of group method of data handling (GMDH). Each node in the network is very flexible and can carry out polynomial type mapping between input and output variables. But the performances of PNN depend strongly on the number of input variables available to the model, the number of input variables, and the type (order) of the polynomials to each node. In this paper, GA is implemented to better use the optimal inputs and the order of polynomial in each node of PNN. The appropriate inputs and order are evolved accordingly and are tuned gradually throughout the GA iterations. We employ a binary coding for encoding key factors of the PNN into the chromosomes. The chromosomes are made of three sub-chromosomes which represent the order, number of inputs, and input candidates for modeling. To construct model by using significant approximation and generalization, we introduce the fitness function with a weighting factor. Comparisons with other modeling methods and conventional PNN show that the proposed design method offers encouraging advantages and better performance.

A new control scheme is proposed to improve the system performance for discrete-time fuzzy systems by tuning control grade functions using neural networks. According to a systematic method of constructing the exact Takagi-Sugeno (T-S) fuzzy model, the system uncertainty is considered to affect the membership functions. Then, the grade functions, resulting from the membership functions of the control rules, are tuned by a back-propagation network. On the other hand, the feedback gains of the control rules are determined by solving a set of LMIs which satisfy sufficient conditions of the closed-loop stability. As a result, both stability guarantee and better performance are concluded. The scheme applied to a truck-trailer system is verified by satisfactory simulation results.

In unidentifiable models, the Bayes estimation has the advantage of generalization performance over the maximum likelihood estimation. However, accurate approximation of the posterior distribution requires huge computational costs. In this paper, we consider an alternative approximation method, which we call a subspace Bayes approach. A subspace Bayes approach is an empirical Bayes approach where a part of the parameters are regarded as hyperparameters. Consequently, in some three-layer models, this approach requires much less computational costs than Markov chain Monte Carlo methods. We show that, in three-layer linear neural networks, a subspace Bayes approach is asymptotically equivalent to a positive-part James-Stein type shrinkage estimation, and theoretically clarify its generalization error and training error. We also discuss the domination over the maximum likelihood estimation and the relation to the variational Bayes approach.

This paper deals with the use of neural network rule extraction techniques based on the Genetic Programming (GP) to build intelligent and explanatory evaluation systems. Recent development in algorithms that extract rules from trained neural networks enable us to generate classification rules in spite of their intrinsically black-box nature. However, in the original decompositional method looking at the internal structure of the networks, the comprehensive methods combining the output to the inputs using parameters are complicated. Then, in our paper, we utilized the GP to automatize the rule extraction process in the trained neural networks where the statements changed into a binary classification. Even though the production (classification) rule generation based on the GP alone are applicable straightforward to the underlying problems for decision making, but in the original GP method production rules include many statements described by arithmetic expressions as well as basic logical expressions, and it makes the rule generation process very complicated. Therefore, we utilize the neural network and binary classification to obtain simple and relevant classification rules in real applications by avoiding straightforward applications of the GP procedure to the arithmetic expressions. At first, the pruning process of weight among neurons is applied to obtain simple but substantial binary expressions which are used as statements is classification rules. Then, the GP is applied to generate ultimate rules. As applications, we generate rules to prediction of bankruptcy and creditworthiness for binary classifications, and the apply the method to multi-level classification of corporate bonds (rating) by using the financial indicators.

This paper proposes homogeneous neural networks (HNNs), in which each neuron has identical weights. HNNs can realize shift-invariant associative memory, that is, HNNs can associate not only a memorized pattern but also its shifted ones. The transition property of HNNs is analyzed by the statistical method. We show the probability that each neuron outputs correctly and the error-correcting ability. Further, we show that HNNs cannot memorize over the number,, of patterns, where m is the number of neurons and k is the order of connections.

In this paper, we present a learning approach, positive correlation learning (PCL), that creates a multilayer neural network with good generalization ability. A correlation function is added to the standard error function of back propagation learning, and the error function is minimized by a steepest-descent method. During training, all the unnecessary units in the hidden layer are correlated with necessary ones in a positive sense. PCL can therefore create positively correlated activities of hidden units in response to input patterns. We show that PCL can reduce the information on the input patterns and decay the weights, which lead to improved generalization ability. Here, the information is defined with respect to hidden unit activity since the hidden unit plays a crucial role in storing the information on the input patterns. That is, as previously proposed, the information is defined by the difference between the uncertainty of the hidden unit at the initial stage of learning and the uncertainty of the hidden unit at the final stage of learning. After deriving new weight update rules for the PCL, we applied this method to several standard benchmark classification problems such as breast cancer, diabetes and glass identification problems. Experimental results confirmed that the PCL produces positively correlated hidden units and reduces significantly the amount of information, resulting improved generalization ability.

The lifting scheme is an efficient and flexible method for the construction of linear and nonlinear wavelet transforms. In this paper, a novel lossless image coding technique based on the lifting scheme using discrete-time cellular neural networks (DT-CNNs) is proposed. In our proposed method, the image is interpolated by using the nonlinear interpolative dynamics of DT-CNN, and since the output function of DT-CNN works as a multi-level quantization function, our method composes the integer lifting scheme for lossless image coding. Moreover, the nonlinear interpolative dynamics by A-template is used effectively compared with conventional CNN image coding methods using only B-template. The experimental results show a better coding performance compared with the conventional lifting methods using linear filters.

This letter studies a simple nonautonomous chaotic circuit constructed by adding an impulsive switch to the RCL circuit. The switch operation depends on time and on state variable through a refractory threshold. The circuit exhibits various chaotic attractors, periodic attractors and related bifurcation phenomena. The dynamics can be analyzed using 1-D return map focusing on the time-dependent switching moments. Using a simple test circuit model typical phenomena are verified in PSPICE simulations.

The application of neural networks to the state-feedback guaranteed cost control problem of discrete-time system that has uncertainty in both state and input matrices is investigated. Based on the Linear Matrix Inequality (LMI) design, a class of a state feedback controller is newly established, and sufficient conditions for the existence of guaranteed cost controller are derived. The novel contribution is that the neurocontroller is substituted for the additive gain perturbations. It is newly shown that although the neurocontroller is included in the discrete-time uncertain system, the robust stability for the closed-loop system and the reduction of the cost are attained.

This paper presents pulse-coupled two bifurcating neurons. The single neuron is represented by a spike position map and the coupled neurons can be represented by a composition of the spike position maps. Using the composite map, we can analyze basic bifurcation phenomena and can find some interesting phenomena that are caused by the pulse-coupling and are impossible in the single neuron. Presenting a simple test circuit, typical phenomena are confirmed experimentally.

The paper first researches the properties of neural networks in the framework of the dual linear programming theory, then discusses the variation range of a Hessian matrix associated to dual linear programming problems. By means of eigenvalues method, a Lipschitz constant based formula for determining the algorithm step-size is presented. Two examples are given to show that the proposed formula is efficacious.

Nonminimum phase systems are difficult to be controlled with a conventional PID-type controller because of their inherent characteristics of undershooting. A neuro-controller combined with a PID-type controller has been shown to improve the control performance of the nonminimum phase systems while maintaining stability. In this paper, we apply a multiobjective evolutionary optimization method for training the neuro-controller to reduce the undershooting of the nonminimum phase system. The computer simulation shows that the proposed multiobjective approach is very effective and suitable because it can minimize the control error as well as reduce undershooting and chattering. This method can be applied to many industrial nonminimum phase problems with ease.

In this paper, an adaptive bound reduced-form genetic algorithm (ABRGA) to tune the control points of B-spline neural networks is proposed. It is developed not only to search for the optimal control points but also to adaptively tune the bounds of the control points of the B-spline neural networks by enlarging the search space of the control points. To improve the searching speed of the reduced-form genetic algorithm (RGA), the ABRGA is derived, in which better bounds of control points of B-spline neural networks are determined and paralleled with the optimal control points searched. It is shown that better efficiency is obtained if the bounds of control points are adjusted properly for the RGA-based B-spline neural networks.

Scientific computations for diffusion equations and ANNs (Artificial Neural Networks) are data intensive tasks accompanied by heavy memory access; on the other hand, their computational complexities are relatively low. Thus, this type of tasks naturally maps onto SIMD (Single Instruction Multiple Data stream) parallel processing with distributed memory. This paper proposes a high performance acceleration processor of which architecture is optimized for scientific computing using diffusion equations and ANNs. The proposed architecture includes a customized instruction set and specific hardware resources which consist of a control unit (CU), 16 processing units (PUs), and a non-linear function unit (NFU) on chip. They are effectively connected with dedicated ring and global bus structure. Each PU is equipped with an address modifier (AM) and 16-bit 1.5 k-word local memory (LM). The proposed processor can be easily expanded by multi-chip expansion mode to accommodate to a large scale parallel computation. The prototype chip is implemented with FPGA. The total gate count is about 1 million with 530, 432-bit embedded memory cells and it operates at 15 MHz. The functionality and performance of the proposed processor is verified with simulation of oil reservoir problem using diffusion equations and character recognition application using ANNs. The execution times of two applications are compared with software realizations on 1.7 GHz Pentium IV personal computer. Though the proposed processor architecture and the instruction set are optimized for diffusion equations and ANNs, it provides flexibility to program for many other scientific computation algorithms.